When you sit down with an AI music generator, the temptation is to toss in a vibe and hope for magic. Sometimes it works. More often, you get a track that almost fits, but not quite: a chorus that never lifts, a bassline that crowds the kick, or a pad that drifts without purpose. The difference between “close” and “keeper” is rarely gear. It’s how you frame the ask. Prompts are the brief, the score sheet, and half the producer’s instinct packed into a few sentences.
I’ve spent years writing briefs for human producers, scoring cues for libraries, and building prompt systems for generative tools. The same truths hold across people and models: precision creates freedom, constraints spark creativity, and context reduces guesswork. If you give a model style, arrangement, tempo, harmony, sound palette, and a hook target, you’ll get material you can work with. Skip those, and you’ll sift for hours.
This guide shows how to structure prompts for genres, moods, and hooks, the three levers that most directly shape musical identity. Along the way, you’ll see tested prompt formulas, workflow tips, and pitfalls I learned the hard way. Use them with your preferred AI music generator, pair them with an AI text generator for lyrics, or bring them into a full-stack workflow with ai writing tools and ai video generator cues. Treat these as starting points, then adapt to your ear.
What a useful music prompt actually contains
A music prompt is more than “lofi chill beats” or “epic orchestral.” Those tags describe bins, not songs. A strong prompt delivers five layers: genre markers, mood and energy, arrangement and structure, harmonic range, and production details. That sounds heavy, but it fits in a compact paragraph if you keep it concrete. Think “four-bar intro, 16-bar verse, no sub in the verse, snare on 2 and 4, D dorian, 92 BPM” rather than “make it cool.”
When I prompt for a cue, I include references sparingly. If I name artists or tracks, I adjust the request so it becomes a descriptive reference rather than a cloning attempt: “in the spirit of early 2010s deep house, organ bass and swung hats” instead of a band name. For some platforms, using named references is fine, but descriptive references travel better across models.
Finally, decide what role the output plays. Is this a loop pack, a 60-second promo, a full song demo, or a stem bed for ai for marketing? The deliverable affects structure. A product demo might need tight edit points every 4 bars. A songwriting demo should lean into a memorable chorus and a singable topline.
A compact prompt formula that works
Here’s a tight prompt structure I’ve found robust across different ai music generator tools:
Role or use case, genre and era, mood and energy, tempo and meter, key or scale, harmonic movement, instrumentation and sonic palette, arrangement and sections with durations, hooks or motifs, mix and spatial notes, technical constraints.
That looks long on paper. In practice, it’s 3 to 6 sentences.
Example: “Two-minute high-energy synthwave cue for a product teaser, early 80s spirit with modern punch. Confident, neon, forward momentum at 120 BPM in 4/4, A minor natural. I-VI-III-VII rotation with occasional iv borrowed chord, gated tom fills and syncopated bass arpeggios. Instruments: Juno-style poly pads, bright lead with subtle glide, punchy Linn-style drums, chorus bass. Structure: 4-bar filtered intro, 16-bar A, 16-bar B with octave bass and open hats, 8-bar breakdown, 16-bar finale. Hook: 8-note lead motif repeating with rhythmic variation. Mix: wide pads, dry snare, moderate bus compression, no reverb tails past section cuts.”
Most generators handle something like this cleanly. If yours accepts JSON, turn each parameter into a field. If you’re using a text-only interface, keep sentences short and atomic.
Genre prompts that hold up under scrutiny
A genre prompt sets the language the track speaks. Genre is rhythm plus instrumentation plus mix aesthetic. If you ignore one of those, your model will fill in the blanks, and you may not like its guesses.
House and dance
House lives in the pocket of the kick and bass. Ask for power where it matters and restraint where it doesn’t. For classic house: “Club-friendly classic house at 124 BPM, 4/4, D minor. Tight 909 kick, crisp open hats with shuffle, clap on 2 and 4. Bass is warm, short notes, sidechain pump. Piano stabs on offbeats, subtle organ layer. Structure: DJ-friendly, 8-bar intro with filtered elements, 16-bar groove, 16-bar lift with riser and chord expansion, 16-bar drop, 16-bar outro with stripped drums. Hook: simple 2-bar piano riff repeating, slight variations. Mix: mono-source kick and bass below 120 Hz, wide hats and stabs, gentle tape saturation.”
For tech house, remove the piano, use darker percussion, and call out the drum accents. For melodic house, specify long pads, arps, and a chord progression with stepwise motion.
Hip-hop and trap
Good hip-hop prompts control space. Too many models oversaturate the midrange. Ask for air and groove. “Minimal trap beat, 142 BPM halftime feel, F sharp natural minor. Sparse 808 bass with clean glide notes, kick supports 808 rather than stacking, snare sharp and short, hi-hats with triplet bursts and occasional 1/64 stutters. Texture: ethereal ambient pad, one melancholy bell motif, vinyl crackle at low level. Structure: 2-bar intro with no drums, 8-bar A, 8-bar B with hat switch-up and counter-melody, 4-bar drop, repeat. Leave space for rap vocal 200 Hz to 3 kHz, avoid dense mid pads. Hook: simple 4-note bell motif.”
For boom bap, slow down, add swing, and specify sample chops. If you want drill, call out the sliding 808 patterns, triplet hi-hats, and minor-key strings or choir stabs.
Pop
Pop prompts benefit from specificity about chorus lift and a catchy topline target. “Modern pop with indie shimmer, 100 BPM, B major, 4/4. Verse sparse with palm-muted guitar and soft kick, pre-chorus adds toms and rising synth, chorus opens wide with sidechained synth examples of ai-generated illustrations pad and layered claps. Chords: I - V - vi - IV with a 2-bar pre-chorus on ii - IV - V. Hook: 7-syllable vocoder ‘oh-oh’ line, easy to whistle. Mix: radio-bright top end, controlled low end, polite reverb, slap delay on lead only in chorus. Bridge: half-time feel and temporary parallel minor color.”
If your generator supports lyrics or vocal guides via ai text generator tools, supply syllable counts and rhyme frames. Pop tolerates exactness. Vague pop prompts produce mush.
Orchestral and cinematic
Orchestral prompts require detail about register, density, and dynamics. “Cinematic adventure cue, 120 BPM, 4/4, in D major with occasional modal interchange to D mixolydian. Section map: 8-bar noble theme in horns and violins, 8-bar development with woodwind runs and snare, 8-bar triumphant restatement with choir, 4-bar stinger. Strings: spiccato ostinato at eighth notes, violas and celli doubling, basses on root-fifth. Brass: horns carry melody, trumpets for accents only. Percussion: piatti swells into transitions, snare rolls, low taiko doubling downbeats in B section. Dynamics: mf to ff, no overcompression, space of a concert hall.”
Too many models overstack brass and reverb. Call for articulation types and limit layers per section.
Lofi and ambient
If you want lofi that isn’t a smeared lullaby, constrain noise and movement. “Warm lofi, 84 BPM, 4/4, E dorian. Rhodes main voice with gentle tremolo, nylon guitar accents, soft brushed snare and kick with sidechain for light pulse. 2-bar bass pattern with occasional passing tone. Vinyl noise at -24 LUFS relative, flutter 0.3%. No crash cymbals. Hook: subtle 4-note upper-voice motif returning every 8 bars. Mix: narrow center for low mids, slow tape delay at 1/4 note with low feedback, gentle bus glue.”
Ambient needs motion and focus, not just pads. Ask for evolving timbres and low-density motifs.
Rock and indie
Guitars are their own grammar. “Indie rock, 148 BPM, 4/4, A major with mixolydian color. Two electric guitars: one jangly sixteenth-note pattern with chorus, one crunchy power chords with light overdrive. Bass locked to kick on downbeats, occasional octave jumps. Drums: tight snare, roomy overheads, no clicky kick. Structure: verse - chorus - verse - chorus - bridge - chorus, each 16 bars. Hook: ascending three-note vocal melody over the IV chord at chorus start. Mix: vocals forward, guitars panned 60/60, parallel drum compression. Leave 2-bar re-intro before last chorus.”
For punk, cut reverb, push tempo to 180, make the snare dry, and shorten sections. For shoegaze, commit to wall-of-sound, specify a wash of reverb and slow modulation, and keep the vocal ghosted.
Prompting for mood and energy without resorting to clichés
Mood sits on top of energy. Calm can be tense, happy can be manic. If you only write “uplifting,” the engine will reach for major chords and crashing cymbals. Try pairing emotional vocabulary with musical correlates.
Instead of “sad,” use “restrained, low dynamic, long release tails, melody stays within a 5-note range, chords minor with occasional major IV for ache.” For confidence, specify “steady eighth-note pulse, balanced low end, strong downbeat emphasis, predictable phrase lengths, no dissonant extensions beyond ninths.”
Energy lives in density, rhythm subdivision, and transient punctuation. A sparse arrangement at 128 BPM can feel calmer than a dense one at 95. If you want high energy without clutter, write “high energy via rhythm, not via layering, keep instrument count under seven, use syncopation and percussive fills to carry momentum.”
I keep a mental map that relates descriptors to musical choices. “Warm” implies rolled-off highs, tube or tape saturation, slower attacks on pads. “Dark” hints at lowpass filters, minor modality, and reverb with longer pre-delay. “Intimate” demands close miking and low reverb tails. Translate adjectives into engineering decisions the engine can actually act on.
Hooks the model can’t ignore
Hooks are anchors. They are short, repeatable, and high-contrast against their surroundings. When you prompt, give the hook a job and a shape. “8-note motif, mostly stepwise with one leap of a fifth, starts on offbeat, repeats every 4 bars with one note changed on bar 4, register G4 to E5.” That level of detail forces a clear idea. If vocal-like hooks are allowed, mention syllable counts or vowel shapes: “hook built on ‘oo’ and ‘ah’ syllables, no dense consonants.”
Don’t leave hook placement to chance. Specify where it first appears, how often it returns, and what supports it. “Motif introduced softly in verse as a texture, full statement at chorus with harmony a third above, stripped down for bridge, final chorus adds octave double.”
If you want multiple hook types, name them: lead hook, bass hook, drum signature. A drum signature might be “snare flam into downbeat every 8 bars” or “three-kick pickup into drop.”
Real-world prompt examples by use case
Sometimes you write for a video, not a playlist. Use-case prompts reduce revision cycles because they give the model the practical edges.
Short-form ad: “Thirty-second punchy funk-pop for a sneaker ad, 120 BPM, 4/4, C minor. Crisp slap bass, tight rimshots, choppy wah guitar. Immediate start, no fade in. Edit points every 4 bars with mini stops. Hook: call-and-response horn stabs at bars 5 and 13. Mix: center bass, horns bright, minimal reverb, strong transient shaping. End with a clean button on tonic.”
Background loop for an app: “Seamless 1-minute loop, calm yet focused, 72 BPM, 3/4, D major. Gentle marimba arpeggios, soft felt piano, light brushed snare on beat 2. No sudden cymbals or risers. Dynamics stable within 3 dB. Hook should be understated: three-note piano rise. Low frequency limited to 60 Hz. Loop points click-free.”
YouTube narration bed: “Mid-tempo chillhop bed at 90 BPM, 4/4, G minor. Simple drum groove with soft swing, thick bass with sidechain, Rhodes and muted guitar. No busy melodies, leave room at 500 Hz to 3 kHz. Section changes every 8 bars via percussion swaps and filter moves. Hook is textural, not melodic.”
Game battle loop: “High-intensity orchestral hybrid, 140 BPM, 4/4, E minor harmonic. Aggressive strings in staccato 16ths, taiko ensemble, distorted synth bass doubling root. Brass in short bursts only. Loopable 60 seconds with strong downbeat. Harmonic cycle i - VI - III - VII. Mix: cinematic, low end controlled, sidechain to avoid brass masking strings.”
Podcast theme: “Fifteen-second theme, confident and friendly, 100 BPM, A major. Funky bass, claps, clean guitar with muted strums, short synth lead doing a 5-note theme. Start with instant hook, no intro. End with tidy button. No lyrics.”
Prompt variations for specific subgenres
When you narrow to subgenres, call out micro-traits. For drum and bass, specify break types. For reggaeton, mention dembow patterns. For afrobeat, define rhythm guitar roles and horn voicings. If you want K-pop sheen, prioritize ultra-tight edits, rhythmic vocal chops, and bright, layered choruses with multiple micro-hooks.
A few examples:
- Drum and bass, liquid: “174 BPM, 4/4, B minor. Liquid DnB with rolling Reese bass, airy pads, and bright piano stabs. Use Amen-inspired break variations with ghost notes, light ride cymbal, and snare at 2 and 4. Lush reverb but keep kick and bass clean. 16-bar sections, soft intros and breakdowns, no harsh neuro growls.” Reggaeton: “96 BPM, 4/4, F minor. Dembow rhythm with crisp snare on 2 and ‘uh’ of 2, syncopated kick, percs including woodblock. Sparse minor-key synth plucks, sub bass minimal. Hook is a 2-bar vocal-like synth riff. Minimal reverb, stereo width via midrange synth only.” Future bass: “146 BPM, 4/4, D major with borrowed vi minor in pre-chorus. Lush detuned supersaw stacks, sidechain pump obvious, hyper-detailed drum fills in transitions. Melodic drop centered on pitch-bent lead. Bright mix, tight low end.”
These differences look subtle in text, but they change the output a lot.
Mixing and space, the hidden part of prompt engineering
Many models default to big reverbs and loud masters. If you want clarity, ask for it. “Bus compression at 2:1, slow attack, medium release, 1 to 2 dB gain reduction. Reverb short for drums, medium plate for vocals, pre-delay 20 ms. Mono low end up to 120 Hz. Master loudness target around -10 LUFS integrated, true peak below -1 dB.”
Space is musical. Tell the engine where to place instruments. “Kick and bass centered, guitars panned 40L/40R, lead mid, pads wide. Percussion toys at 60L/60R.” If your generator accepts stem output, request stems by instrument family. Stems let you fix what the model misread.
Prompting hooks for lyric-driven tracks
If your tool supports a voice generator or translates syllable guides into vocal melodies, give a simple lyric skeleton and meter. I often write a metered line without final words to guide rhythm: “VERSE, 8 bars, 4 stresses per line, last stress anticipates beat 4. PRE, rising stress density. CHORUS, 6 lines, 7 syllables each, end-rhyme ABABCC, vowel focus on ‘ah’ and ‘ee’ to cut.”
Then add the melodic contour: “Chorus melody starts on 5th of the scale, climbs to 6th on line 2, drops to 3rd on line 4, tag ends on tonic with a 2-note pickup.” With that much detail, the engine tends to deliver singable shapes rather than melismatic wander.
For rap hooks, specify cadence and rhyme map: “4 bars, internal rhymes on beats 2 and ‘and’ of 3, end rhymes aaab. Simple syllables, percussive consonants, no long vowels sustained over bar lines.”
A two-list allowance used wisely
Checklist for a strong prompt you can reuse:
- Purpose and length: where it plays, how long, loopable or not. Genre and subgenre with era: house 1998, drill 2020s, synthwave retro-modern. Tempo, meter, key or scale, and a chord roadmap. Instrumentation and sonic palette, including drum kit identity. Structure with bar counts, hook placement, and mix constraints.
Common mistakes that waste time:
- Writing feelings without translating them into musical parameters. Overstuffing the middle: too many instruments, no clear lead. Forgetting edit points or button endings for media use. Letting reverb and delay smear transitions. Ignoring low-end discipline, especially kick versus bass.
How to iterate without losing the thread
Treat your first pass as the sketch. Listen with intent: is the groove there but the sound wrong, or is the harmonic movement off? Change only the piece that’s wrong, and say so. “Keep the exact rhythm and structure. Replace the bass with a shorter, tighter tone, no sustain. Reduce reverb by half on the snare. Increase chorus intensity at the last 16 bars.”
If you need a bolder shift, anchor what worked. “Preserve the hook and tempo. Change the mode from natural minor to dorian. Swap the piano for a plucked synth. New bridge with half-time feel and filtered drums.” The more you preserve explicitly, the less the model will throw away.
Use prompt testing as a practice. Keep a prompt library, tag your successes by genre, mood, and model version, and make notes on what each engine tends to overdo. This becomes your personal ai prompt library for music, similar to how creators organize midjourney prompts or stable diffusion prompts for ai image generation.
Pairing music prompts with visual or narrative prompts
If you work on package content where audio supports visuals, sync the two briefs. A fast-cut edit benefits from more edit points and rhythmic motifs that land on cuts. A slow dolly shot appreciates longer swells. You can even embed cut structure in the prompt: “video has cuts at bars 1, 5, 9, 13, 17, include accents or fills on those downbeats.” For ai content creation that blends ai art generator outputs with sound, match color temperature to timbre: warm visuals often pair with mid-rich instruments like Rhodes and sax; cold palettes pair with glassy FM synths or celesta.
For storytelling prompts that feed both words and music, align arcs. If your ai storytelling outline includes a midpoint reversal, build a contrasting B section or a IV-major lift. If your ai video generator sequence calls for a 12-second slow reveal, ask for a 12-second riser with a modal shift.
Business use: prompts that deliver results for clients
When music supports a brand, capture brand identity in sonic terms. If a brand voice guide says “bold, friendly, modern,” translate that: bold equals solid low end and clear transients, friendly equals consonant intervals and warm upper mids, modern equals crisp top end and clean low noise. If you do ai for business or ai for marketing, write prompts that target brand consistency, then template them. Over time, you’ll build a prompt strategy that saves rounds of feedback.
For SEO-driven content and ai blog writing, pre-scored templates speed production. Assign tones by content pillar: explainer equals focused and neutral beds with minimal hooks; product walk-through equals higher energy, edit points, and brighter mixes. Use ai automation to attach cue templates to content types in your workflow.
Ethical and practical guardrails
Avoid cloning artists. Describe features, not names, unless your tool permits references and your use is licensed. When using public prompts, test for distinctiveness. If you hear identical chord moves and sound choices across outputs, push your prompts toward unusual progressions, scales like dorian or lydian, quirky meters like 7/8 sections, or unusual instrument combos. You want originality with relevance.
Be careful with loudness. Models sometimes deliver hot masters that fatigue ears. Ask for LUFS targets. If you’re mixing later, ask for headroom, like peaks at -6 dBFS. If you plan to distribute or pair with voice-over, measure, don’t guess.
Prompt fragments you can paste and adapt
You can drop these into your own prompts as add-ons.
- Chord movement: “Use I - V - vi - IV in verse, shift to vi - IV - I - V in chorus, pre-chorus ii - IV - V tension.” Rhythm feel: “Swing at 55%, hi-hat patterns vary every 2 bars with one open-hat accent.” Hook architecture: “8-note motif, repeat three times with the fourth iteration altered, end on tonic before downbeat.” Mix framing: “Mono low end to 120 Hz, stereo width increases with section intensity, ceiling true peak -1 dB.” Edit readiness: “Hard stops or cymbal chokes at bars 16, 32, 48 for clean cuts.”
Treat these as modular components in your prompt syntax. Over time you’ll craft a prompt design style that reflects your taste.
From prompt to production: a pragmatic workflow
I run a simple loop. First, I draft a concise prompt using the formula, generate two or three variations, and pick the one with the best groove, not the best polish. Second, I request stems if supported. Third, I do surgical edits: align transitions, trim reverb tails, tune the hook if needed, and tighten the kick-bass relationship. Fourth, I layer a single live element where feasible, even a handclap recorded on a phone, to add human air. Fifth, I master gently or drop the track into a larger ai workflow where a mixing assistant handles polish.
This flow beats trying to get the perfect file from a single run. Models are fast. Iterate for content, not finish. Finish in your DAW or with a specialized tool.
Troubleshooting weird outputs
If the output feels aimless, you probably didn’t lock the structure tightly enough. Add bar counts and cadence markers. If it’s harmonically dull, provide a progression with one color chord. If the mix is hollow, ask for midrange focus and reduce over-wide stereo. If percussion is over-busy, restrict the number of percussive layers and specify where fills belong. If the hook disappears, promote it in the prompt by naming its role and recurrence.
Edge cases come up. Some models struggle with odd meters or key changes. If you need those, try embedding them as sections: “bridge switches to 3/4 for 8 bars, then back to 4/4” rather than asking for fluid polymeter. If the engine overcompresses, explicitly ask for dynamic range and lower integrated loudness. If a generator leans toward glossy pop gloss in every genre, lean harder on instrument identity and mic technique language.
Using prompts across creative domains
Music prompts improve when you think like a multi-disciplinary creator. The habits you learned crafting ai image prompts for midjourney prompts or stable diffusion prompts carry over. In images, specifying lens, lighting, and composition yields consistent results. In music, specifying mic character, room size, and arrangement does the same. If you use ai prompt examples in a visual moodboard, convert those words into sound adjectives and engineering equivalents.
The same goes for ai copywriting and songwriting. Strong headlines are hooks. They’re concise, rhythmic, and repeated. Use ai storytelling prompts to outline an arc, then build your sections to mirror that arc. If you work with ai tools for creators across audio, visual, and text, unify your prompt library and share a prompt guide with your team so the sonic identity ties back to the brand or project story.
A few complete, ready-to-use prompts
Synthwave anthem: “Two-minute synthwave anthem for a product reveal, 120 BPM, 4/4, A minor. 80s palette: analog polysynth pads, arpeggiated bass with mild distortion, Linn-style drums with gated reverb on toms. Chords: i - VI - III - VII. Structure: 4-bar riser intro, 16-bar A with pad and bass, 16-bar B with bright lead and open hats, 8-bar breakdown with filtered bass only, 16-bar finale doubling the lead in octaves. Hook: 8-note lead motif, stepwise with one leap to the 6th. Mix: wide pads, dry snare, bus compression subtle, true peak -1 dB. Edit points at bars 16, 32, 48.”
Afrobeats groove: “Afrobeats-influenced pop, 102 BPM, 4/4, G major. Dotted rhythms in drums, soft kick, rim on 2 and ‘and’ of 3, congas subtle. Guitar does interlocking palm-muted patterns. Bass bouncy, note length short. Chords: I - vi - IV - V. Hook: call-and-response between vocal-synth chop and guitar lick. Structure: verse 16 bars, pre 8 bars with rising percussion, chorus 16 bars, post-chorus 8 bars with instrumental hook. Mix: warm and dry, stereo percussion details.”
Cinematic tension bed: “Investigation tension bed, 110 BPM, 4/4, C minor. Low ticking percussion, soft pulses, string harmonics and col legno, pings from prepared piano. Harmonic center mostly on i with chromatic movement in the bass. Build intensity every 8 bars by adding layers, no giant hits. Hook is a 3-note descending motif on muted piano. Keep dynamic range, integrated loudness around -18 LUFS, no sub boom.”
Lofi study loop: “Seamless 1:30 loop for studying, 78 BPM, 4/4, D dorian. Felt piano, soft vinyl texture, brushed snare, upright bass with light tape saturation. Chords: ii - IV - I - V in dorian color. Hook: gentle 4-note piano figure every 8 bars. Minimal high-frequency content above 10 kHz. Low reverb, slow chorus on piano, loop points clean.”
Indie pop with lyrical chorus: “Indie pop song demo, 112 BPM, 4/4, E major. Verse sparse with muted guitar and kick, pre adds tom build, chorus wide with jangly guitars and claps. Chords: I - V - vi - IV, pre-chorus hangs on ii - V. Hook: 6-syllable chorus phrase starting on scale degree 5, melody climbs to 6 then resolves to 3. Mix radio-bright, vocals forward, chorus vocals double-tracked.”
From experimentation to a reusable system
Treat prompt engineering for music like any craft. The first dozen tries feel clumsy. By the 30th, you’ll hear patterns and learn what your chosen model likes. Keep a lightweight prompt generator of your own, even a simple text template with blanks for tempo, meter, key, chord roadmap, and hook style. Tag your favorite outputs and their prompts, just as visual artists keep ai prompt examples for reference. If you’re collaborating, store everything in a shared ai prompt guide so your team can move faster and avoid repeating mistakes.
The aim isn’t to make the model do everything. It’s to get a draft that respects your taste and saves time. When the groove, harmony, and hook land, polishing is a joy. When the foundation is wrong, you’re fighting uphill. Better prompts fix that, whether you’re crafting a cinematic bed, a viral short, or a brand theme that will loop a million times.
Ultimately, the best prompts read like a producer talking to another producer. Clear, concrete, and musical. Give the engine the map, and it will take you somewhere worth hearing.